(A) study on flexible decomposition approaches to decentralized resource allocation

Abstract

In designing an efficient decentralized resource allocation mechanism, decomposition theory is widely used as an organizing device. Recent developments in paralled processing technology for large-scale computation revive the research interests in decomposition methods. We pay particular attention to the implementability issues of the decomposition-based coordination mechanisms. In this study, we suggest two kinds of new decomposition method to enhance the implementability. To analyze the conflicts between the center and the civisions, a new analysis framework is introduced. Most of decomposition methods utilize either primal (resource) information or dual (price) information, not both, as a control signal from the center. The implied information exchange process between the center and the divisions has been assumed to be in a predetermined fashion regardless existing informational structure. To resolve this rigidity, we propose a mixed decomposition method, where the center can select an arbitrary combination of price and quantity controls for common resources to each division. This is realized in two ways; i) two master problems are introduced so that one generates price gsignals and the other quantity signals, and ii) one hybrid master problem deals with both price and quantity information. The suggested methods are well justified in terms of the flexibility in information flow pattern choice (some prices and other quantities). An organization is often considered as composed of a master or headquarters that is solely responsible for an optimal and feasible resource allocation and selfish divisions giving no attention to the rest of the system. Nevertheless, in the context of conflicts between the center and the divisions, there has been virtually no proper model to describe the real world situation within the decomposition framework. Such questions can be answered by a variety of plausible behavioral assumptions in the decentralized decision making procedure....